Dan Rockmore teaches mathematics and computer science at Dartmouth. His Weekend Essay asks why we treat physical training and intellectual training as if they obeyed different laws — as though the mind might get strong without being made to struggle.

The gym supplies the comparison:

  • Nobody in a spin class would dream of outsourcing the pedaling, even though a car or e-bike covers the same distance more easily
  • Forty-five minutes of deliberate discomfort is accepted because it builds a future self — the exercise is the point
  • The classroom runs on proxies too: problem sets, close readings, research papers were never the skill itself, they were the bench press

What changed is that machines can now do the reps. An early-2026 AI tool called Einstein claimed to plug into Canvas, ingest homework, complete and submit it on a student’s behalf — the product vanished almost as fast as it appeared, but the outrage it provoked was misplaced. The question was never “how dare a machine do the homework.” It was “why was the homework mechanical enough for a machine to do it?”

Rockmore’s answer is that higher education spent years teaching students the point of a class was the credential. If the grade is the goal, and a machine can get the grade, then the machine is not a shortcut. It is efficiency.

What the gym has that the classroom usually lacks is legible feedback:

  • You feel the burn, see the weight go up, watch the resistance knob as a continuous private metric
  • Wrestling with a hard text or idea can take days or weeks before the clouds clear
  • The Greeks ran athletics and philosophy in the same building — the gymnasion, where Plato taught; the German gymnasium kept the name and dropped the athletics

His examples of classrooms that engineer the equivalent loop: a colleague who lets students retake exams until concepts click, treating mastery the way a coach treats a skill; the old S.R.A. Reading Laboratory with its color-coded cards, “a gym for reading”; and Kyle Booten’s book Gyms, a set of computational writing tools built to make writing harder rather than easier — one demands rare words, another is a literary CrossFit box for strange sentence constructions.

The practical conclusion is not anti-AI. Let the machines generate practice problems, flag errors, and grade what is beyond discussion — that only sharpens what teachers are actually for: drawing out the quiet student, asking the question nobody thought to ask, building a community out of shared difficulty. Institutions already know how to hold that space open. Dartmouth has no athletic scholarships, and the student Rockmore watches keeping climbing the high-dive ladder anyway, because he has a team and a lane.

The most valuable part of the athletic experience is what is left when the athlete no longer takes the field.